Design considerations for parallel power

Author:
Patrick Le Fèvre, Ericsson Power Modules

Date
10/26/2014

 PDF
Multicore hardware can provide load balancing and resiliency

Hardware manufacturers are taking advantage of increasing integration and improvements in technology to pack more functionality into a smaller space. In servers and communications switches they are applying the benefits of multicore hardware to provide load balancing and resiliency as well as technologies such as voltage and frequency scaling to fine-tune energy usage as compute demand changes. This is leading to changes in the way that power-delivery subsystems are designed.

There is an increasing demand for fault tolerance and redundancy as more processors are consolidated into a single box. At the same time, the power-delivery subsystems have to be able to source high levels of current during peak loads, which may be short-lived.

The power and current levels provided by today’s intermediate bus converters continue to increase but in these densely packed servers, there are current demands that are even higher than the output of a single brick. In such cases, the power system designer may consider the option of paralleling two or more units (see Figure 1). Paralleling brings further benefits in terms of resilience through the implementation of ‘n+1’ redundant designs. In these configurations, an additional board-mounted power supply is provided in addition to the number needed to power the system at full load so that the overall energy subsystem can provide uninterrupted system availability in the event of a failure in one of the modules.

Click image to enlarge

Figure 1 Two power DC/DC converters interconnected in parallel offering higher power and optional redundancy

There are other reasons for using a parallel configuration, such as employing two or more lower-power modules to make it easier to fit the power-delivery subsystem into a thinner space where a single module would be too tall – a key consideration in densely packed blade servers. This approach could also be used to distribute the thermal load over a larger board area. A further possibility is to use different combinations of a small selection of modules to reduce stocking requirements for manufacture and for spares.

Applying parallel techniques to power delivery cannot be done automatically with any type of module. Unless the manufacturer claims that they are capable of being paralleled, it should be assumed that a parallel connection is not possible. One reason is manufacturing tolerances. Even for modules that carry the same part number, the output voltage of each converter will be slightly different. When paralleled, the converter with the highest output voltage will source the most current and can go into its current limit mode.

If the converter is designed with either a fold-back or shutdown type of current limit, it can turn off or lock up and the system will not be be able to operate. Even with a constant-current type of overcurrent implementation, the module with the highest output voltage will operate at full load, which reduce its level of reliability compared to other converters in the paralleled connection because of the thermal and other stresses on the circuitry.

Converters that are designed for paralleling generally use one of two types of current sharing methods to insure that any one converter is not overloaded. One is slope compensation, where each converter is designed is designed to automatically adjust its output voltage downward as its current increases. This helps each converter share the total output current more fairly.

A second approach is active current sharing, a more complex design normally used for converters with higher output power ratings. A signal line connects the paralleled converters and actively controls their internal switching operation as a function of the sensed current level in each converter. An alternative to this approach is to employ external integrated circuits to adjust the output of each module to provide a fairer balance of output current.

Paralleling for redundancy introduces further considerations. The configuration needs to be able to supply full power to the load even with the failure of one of the modules. Further, the failed module must not disrupt the voltage output of either the input or output rails.

The output bus is typically protected by the use of an ORing circuit in series with the output of each power module. A low-voltage drop device is used to minimize the loss in overall efficiency. In the event of a module failure, this ORing circuit will become back-biased and prevent current from the other modules from flowing into the inactive module. Any common alarm or control circuits on the secondary side of the paralleled power modules should be either isolated or implemented with high impedance components in order to minimize leakage currents.

Some of the failure modes of a power module result in a short across the input to the module. If no precautions are taken, this short will draw excessive current from the DC intermediate voltage input bus and depress the DC input voltage, disrupting power delivery to the remaining modules. This condition is prevented by using a fuse at the input to each power module in a redundant configuration. The fuse should be sized to conduct the worst-case input current to the module at the lower extreme of the input bus voltage.

Although there are important architectural considerations to make in the implementation of parallel power, the rise of digital converter control is improving the ease of use of these techniques. Digital control makes it possible to use software to configure the converters within a system rather than pin-header or resistor settings. The software can perform checks to ensure that incompatible settings are not made and help optimize overall system performance.

An example of software control is Ericsson’s Digital Power Designer (see Figure 2), which provides a graphical toolkit for configuring multiple converters within a system. Current sharing can be activated by simply creating a rail definition and adding parallel devices to it. When paralleling, modules will be configured automatically in the Power Designer, simplifying the operation and configuration.

Click image to enlarge

Figure 2 Ericsson Power Designer includes a set of tool simplifying configuration and power system optimization

The same software also provides the user with a way of tuning output filters for optimum performance at design time, taking into account the additional considerations required for parallel-power configurations (see Figure 3).

Click image to enlarge

Figure 3 Simulation tool makes possible to optimize filtering for optimal performances

By combining advanced, parallel-ready power modules together with easy-to-use software, OEMs can be sure of providing resilience and reliability for their increasingly densely packed servers, switches and other high-performance systems.

Ericsson Power Modules 

 

 

 

 

 

 

Related articles

 A Miniature Stretchable Pump for Next Gen of Soft Robots
 Current Sense Chip Resistors Include a 0.3 Milliohm Value
 Enhancing AI Quality by Moving Beyond the Quantitative